Urban walkers seeking guidance online typically learn how long it will take to get from Point A to Point B, but not how beautiful their journey will be.
Researchers at Yahoo Labs are working to change that. They’ve published a paper describing a preliminary attempt to identify the most beautiful, most quiet and happiest ways to walk within a city. They tapped into people’s opinions about the beauty of the places around them, as expressed in their votes on online photos or the tags they used on the photos they uploaded. The result — touted in several recent articles and due to be presented at a conference in Chile in September — isn’t for walkers with limited time. But for those who can spare a few extra minutes for a better urban experience, these routes aspire to provide a more pleasant pathway.
The technology world usually harnesses big data for efficiency and speed, but this project is a pleasant exception: It uses millions of online photos to find the road less traveled, or the one that should be traveled more, if people only had the time.
Subjective ideals such as beauty or happiness are devilishly difficult to define, let alone optimize; not only is beauty in the eye of the beholder, but it can vary by time of day or year, by weather, and by time since the last trash pickup. Given these constraints, I wanted to see how well the researchers had done. Fortunately for me, they’ve published a full set of routes connecting one pair of places in London, where I live. Testing their project would mean taking walks from the Euston Square underground station to the Tate Modern art museum, about two and a half miles apart — as the busy walker walks, not as the crow flies — on opposite sides of the River Thames. I also threw in competing routes suggested by map apps from Google, MapQuest, Bing, Apple, Yahoo and Nokia’s Here.com. I’ve compiled the routes I walked on this Google Map.
I set out on a couple of uncharacteristically warm and sunny days last week. I didn’t walk each route separately. Where two or more routes converged, I sometimes walked that section only once. I nonetheless logged more than 26,000 steps and 12 miles, according to my Fitbit step tracker, stopping frequently to record my impressions of each section, and to log any photos I’d taken.
My verdict: The researchers have done reasonably well.1 The routes their algorithm suggested certainly were more beautiful, quiet and happiness-inducing than the route they identified as the shortest. I’ve explored London often in my two and a half years living here, but these jaunts helped me discover several new pleasant streets, gardens and urban oases.
My walks also showed just how tough it is to optimize for factors other than distance. Construction that may have started after the routing work was done made streets much less beautiful or quiet; maybe they’ll be prettier than ever when the projects are completed. The most beautiful sites or parks often are surrounded by busy roads that are tough to cross. And the quietest pedestrian-only streets often filled with pubgoers spilling out of the buildings on sunny evenings, something that either lends a walk conviviality or hassle, depending on your mood.
I also found that the mainstream routing services deserve more credit than they get. They don’t all suggest the shortest path, and some recommend rewarding deviations.
My two favorite stretches were on routes suggested by MapQuest and Bing. MapQuest took me through the busy Covent Garden area on a partly car-free path. And both MapQuest and Bing had me cross the Thames on the scenic Waterloo Bridge, with St Paul’s Cathedral visible to the east and Big Ben, the London Eye and Parliament to the west. I finished the walk along the South Bank, a pedestrian-only riverside walkway passing by theaters, food vendors and other walkers, some deep in conversation. One woman was asking her companion, “Why are we here?” I think she meant it in the philosophical sense, not the “we’re lost” sense.2
The MapQuest route was significantly longer than other routes suggested by mapping sites — 2.51 miles compared to as little as 2.3 miles for Google’s main route — suggesting MapQuest already is looking beyond speed. And that result wasn’t a fluke. I tested 16 routes across five mapping services — MapQuest, Google, Yahoo, Bing and Here — connecting four prominent underground stations north of the Thames, including Euston Square, with four tourist attractions south of the river, including the Tate Modern.3 MapQuest’s routes were the longest in distance, on average 7 percent longer than Yahoo’s.4
MapQuest spokeswoman Jennifer Asbury said the company doesn’t optimize for anything but speed. “Our pedestrian algorithm is shortest route, as that’s what consumers have requested most,” she said.
Microsoft, which owns Bing, provided a statement that said, “The priority for us is helping people get from point A to point B in an efficient way. If someone is looking for a more picturesque route, we offer Streetside imagery in Bing Maps so they can virtually walk the route prior to their first step.” A Google spokesperson declined to comment, and the other mapping providers didn’t respond to my questions.
One of my least favorite parts of any walk came on what the Yahoo Labs authors tagged the shortest route, as well as the most beautiful one. It was Kingsway, full of notable buildings but also of traffic. Kingsway is perilous to cross, with the sidewalk — called pavement or footway here — often blocked by commuters queuing to enter the busy Holborn underground station. Kingsway was specifically mentioned in Yahoo’s research paper as “unpleasant” and “heavily trafficked.” Yet none of the other routing programs I used suggested walking on Kingsway.5
London presents unique routing opportunities and challenges. Newer cities’ right-angled grids make them easy to navigate and hard to find a routing edge. In central London, few streets keep their same name and direction for more than a half-mile, and parallel to many of the biggest streets run lanes, alleys, closes, courts and gardens too small to allow cars, or to make it onto some printed maps.
The researchers — Daniele Quercia and Luca Maria Aiello from Yahoo Labs in Barcelona, and Rossano Schifanella from the University of Torino — started by divvying up central London into 532 walkable cells, each a square of 200 meters by 200 meters. Then they gathered photos of street scenes from Google Street View and Geograph, a volunteer project to gather photos of every square kilometer in the U.K. and Ireland. They showed these photos side by side to thousands of visitors to UrbanGems.org, a site set up to gather data for the project. These raters — a third of whom live in London, and another third elsewhere in the U.K. — said which of a pair of photos depicted a place that was more beautiful, or made them more happy, or that they found more quiet. Then these ratings were aggregated and used to map routes that optimize for these traits.
This process doesn’t scale well to other parts of London, let alone other cities worldwide — it requires thousands of people to rate photos online. So the researchers also tried another method: They gathered Flickr photos of places within the London area they were studying, along with words used to tag those photos. Then they cross-referenced the tags with the Linguistic Inquiry Word Count, a dictionary of 2,300 commonly used English words, grouped into 72 categories. The researchers then correlated each category with their findings from photo-raters to find which kinds of tags matched up with beauty, happiness and quiet. This produced a formula they can use to derive walking routes anywhere in the world that’s sufficiently documented and tagged by Flickr users. For the purposes of the paper, they provided a fifth London walking route, also optimized for beauty but using their Flickr method.6
The researchers also validated their work by asking 30 Londoners to assess the first four walks. They found each walk was rated at least as highly on its optimized attributes as all the others were; in other words, no route scored higher for happiness than the happy route did. The members of this small panel didn’t assess the routes, like I did, by walking them. They merely studied them on maps.
By relying so heavily on third-party photos, the research may have been led astray. “The visual nature of photographs might result in more attention being paid to photogenic areas that are not necessarily better urban areas,” said Eduardo Cachucho, who helped develop Dérive, a smartphone app that encourages urban exploration, in response to emailed questions. “Nevertheless, it’s great to see an advancement like this.”
“Their work is laudable and I think it’s a very interesting idea to be able to fine-tune navigation beyond just efficiency,” said Justin A. Langlois, research director for Broken City Lab, which developed the Drift app, another urban-exploration app which, like Dérive, was mentioned in the Yahoo Labs research paper.
The paper remains much closer to an interesting idea than a full-fledged application. And these routes aren’t fully described in the paper. The researchers’ maps do not include turn-by-turn directions or zoomed-in insets. I did my best to transcribe these maps onto the paper maps I was using to guide me. One oddity was that both the starting and ending points on the maps looked off. I followed the maps right into the central courtyard of University College London Hospitals’ location near Euston Square, and finished near the staff entrance to the Tate Modern, rather than the main entrance on the banks of the Thames.
The maps also prescribed diversions — go up one road or bridge, then come back. Sometimes these made perfect sense, like when the beautiful route suggested walking halfway across the Millennium Bridge and back to the Tate Modern. The pedestrian-only bridge affords prime views of St Paul’s. At other times, though, these diversions looked like mistakes, such as when the supposedly shortest route included a loop around a rectangular block, or going down one street only to double back.
The shortest route also included two hair-raising encounters with traffic within a couple of minutes. First it led me to cross busy Aldwych at an intersection with no pedestrian crossing, a sprint that company lawyers and my editor probably wouldn’t have approved if they’d known about it. Then, as I approached the Thames, I had to cross Temple Place at Surrey Street, another intersection with no pedestrian crossing and lots of traffic. I contacted Yahoo Labs to arrange an interview about the project but the researchers weren’t available for comment.7
Some challenges can’t be solved without full personalization and customization.8 Some routes work best only at certain times of day. Parts of central London that are saturated with pedestrians during rush hour are quiet and pleasant on weekends. Theatergoers pack the Covent Garden area before curtains are raised, then file inside and leave peaceful streets behind. Many of the routes took me past, but not through, pleasant gardens that lock up at night. Some walkers’ happiness depends most on passing facilities such as toilets or, in my case, a place to sit down to record notes. Also, some beautiful structures reward only walkers who tend to look up.
Or, those who go up and look down. One of the highlights of my walks came right at the start, in that hospital courtyard. An older gentleman smoking a cigarette saw me taking photos and told me there was a great view from the top of the building. I asked if he worked at the hospital. No: “I’ve been here too often for my own good.” We got into the same elevator and he asked where I was from. He was from Kentish Town, nearby in London. He said he’d been to New York, but it rained all weekend. I apologized on behalf of my native city. He got off at 13, in the adult inpatient section. I went on to 16 and enjoyed the view of London, through a panoramic window near the elevator, looking south to the river and beyond.
No app could predict such serendipity; otherwise, it wouldn’t be serendipitous. But walkers who rate their experience as they march around a city could build up a data set that could capture the most magical route between A and B. Until that becomes a reality, or at least until the Yahoo Labs work is fine-tuned, I’ll be checking MapQuest and Bing more often before setting off on walks around London.